Why create overrides?

A predictive model is based on historical data: it analyzes past data to predict the future. 

In the Customer fit model, the segment of a lead (very good, good, medium, low) is defined by the model according to the result returned by historical conversions. However, you may want to override the model to force the segmentation of new leads based on one or more traits. Here's why:

• You have not historically converted a lot of Enterprise companies but you are going up market and want to make sure Enterprise leads are routed to Sales.
• You have converted some small companies in the past but you do not want to waste your Sales team skill on companies with less than 50 employees.  
• There isn't clear evidence from your data about which job, roles and titles are the best but you want to boost some specific job titles based on your Sales team feedback. 

For example, your company has not converted a lot of leads from Enterprise companies, therefore the model "naturally" scores them medium because your historical data say these leads don't convert well. However, you still want to categorize these leads as "very good" for your Sales team to go after. To do so, you would apply an override such as "If the company size of the lead is more than XXXX, then it should be scored very good" [regardless of what the historical data says].

 

Which overrides does Madkudu recommend?

⏩ Follow the tips in this article: A few tips to create relevant overrides

 

How to add an override to a live model?

Pre-requisites 

• You have the permissions of the Architect or Admin role
• You know what a Computation is

Step 1: Duplicate the live model

1. Go to the Data Studio (studio.madkudu.com) 
2. Duplicate the model marked as "live" (live models cannot be edited directly)
3. Pick a name for the duplicated model

Step 2: Create the override

1. 1. Model > Overrides
2. Click on Create new rule
3. Select Form mode
    
   
   • Select Advanced mode only if you need logic that is not supported by the form mode. You'll need to use SQL conditions like you would in a WHERE statement. 
4. Select the Computation, condition, value, rule and segment to create an override. An override is made of a condition ("IF xyz") + a rule on segmentation ("THEN classify lead as low/medium/good/very good")
5. Check the box Case insensitive to make the override work regardless of upper/lower cases in the value. 
6. Click on AND to add a condition. If you need an OR condition, just create another override
7. Click on Save.

Step 3: Assess the impact of new overrides

You want to make sure your newly created override is not degrading the performance of the model by boosting or downgrading too many leads. 

 Visualizing the impact of the newly created overrides over your existing model:

This is useful if you want to check how your new override would impact current model performance.

You'll have open two performance charts: one from your live model with existing overrides, and the other from your duplicated model with your newly created overrides.

1. Open the live model in your browser, head to the Review > Performance tab.

2. Open the same page of the duplicated model where you just created an override. 

Live model without new overrides

Duplicated model with new overrides

By comparing the charts, one can see that the duplicated model predicts 90% of conversions vs 75% for the live model. It also does a better job at discriminating the "good" vs medium "segments".

This comparison is made on the validation dataset (which displays actual conversion rates over your historical sample).

Visualizing the impact of all overrides vs no overrides (training dataset):

This is useful if you want to check how all your overrides (including the newly created one) impact performance compared to no overrides.

1. Click on Overrides impact analysis

2. Open the Thresholds section in another tab. You can visualize the impact of all overrides combined by comparing the two tabs:

In the above example, one can see that deploying all the overrides doesn't improve much performance: with overrides (right chart) 36% of leads [good+very good] predict 84% of conversions. Without overrides, 32% of leads predict 76% of conversions. But it will bring more lead volume to the good and very good segments, so it's a nice add if you're looking to send your sales team more qualified leads.

This comparison is made on the training dataset (which is a rebalenced sample of your historical leads).

Step 4: Deploy new overrides to your live model

Keep these considerations in mind when making your duplicated model live:

1. All your Leads, Contacts, or Accounts will get a new score that includes your override's logic with the next batch scoring. This means all scores should be updated within the next 4-12 hours. You can learn more about update times in this article.
2. Adding, editing, or deleting overrides that increase or decrease prospect scores may trigger automated workflows in your CRM which are based on the customer fit or lead grade score (like your MQL workflow). 

When you are ready, go to the Deployment tab 

• If the live model you are editing is flagged "Live as Standard", click "Deploy model as Standard Fit".
• If the live model you are editing is flagged "Live as Multi-fit", follow this process.

 

F.A.Q 

In which order are my overrides taken into account?

Don't panic if you see your overrides moving around! Madkudu's interface rearranges overrides according to how the model takes them into account, so that you always know which ones get executed first.

More specifically, override priorities follow this flow, with rules being applied one at a time:

1. Should be
   1. Low
   2. Medium
   3. Good
   4. Very Good
2. Should at most be
   1. Low
   2. Medium
   3. Good
   4. Very Good
3. Should at least be
   1. Very good
   2. Good
   3. Medium
   4. Low

Examples:

• If a lead qualifies for a should be low rule, no further rule can promote them back to medium, good, or very good.
• A should at least be rule can only upgrade the segment of a lead up to the limit defined by a should at most be rule (if applied).

Why are downgrading overrides applied first?

Madkudu considers it a greater risk to send your Sales a lead with a downgrading attribute (being a spam, being a student...), even if they have other upgrading attributes (belongs to a large company, is in the target industry...). That's why downgrading overrides are applied first, in order to maximize lead quality and build trust in the model across teams.

What happens if two overrides have exactly the same priority level?

When you create an override with the same level of priority as an existing one, the existing one has, by default, higher priority.

Remember you can always trust the interface to tell you which override will be applied first. If an override is listed above another, that means it has higher priority.

I don't see my CRM field in the picklist, what should I do?

The picklist does not directly contain your CRM fields. It contains Madkudu computations. Some computations map directly to your CRM fields (they just have a similar name), some others come from enrichments sources, and some others are calculated from multiple attributes.

To create a computation from your CRM field, please follow this article.

I created a computation but I don't see it in the picklist, what should I do?

Make sure you have released the computation, following Step 2 of  this article.